From 95b4063de4f418135984b33528c44eeb4f9f4baa Mon Sep 17 00:00:00 2001 From: YiPeng Chai Date: Tue, 19 Mar 2024 10:14:16 +0800 Subject: drm/amdgpu: add interface to update umc v12_0 ecc status Add interface to update umc v12_0 ecc status. Signed-off-by: YiPeng Chai Reviewed-by: Tao Zhou Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c | 9 +++++++++ 1 file changed, 9 insertions(+) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c index f486510fc94c..7006a57277ef 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c @@ -437,3 +437,12 @@ int amdgpu_umc_loop_channels(struct amdgpu_device *adev, return 0; } + +int amdgpu_umc_update_ecc_status(struct amdgpu_device *adev, + uint64_t status, uint64_t ipid, uint64_t addr) +{ + if (adev->umc.ras->update_ecc_status) + return adev->umc.ras->update_ecc_status(adev, + status, ipid, addr); + return 0; +} -- cgit From f27defca68824e8e97218b8816249f258d3d5d32 Mon Sep 17 00:00:00 2001 From: YiPeng Chai Date: Mon, 18 Mar 2024 11:48:07 +0800 Subject: drm/amdgpu: umc v12_0 logs ecc errors 1. umc v12_0 logs ecc errors. 2. Reserve newly detected ecc error pages. 3. Add tag for bad pages, so that they can be retired later. Signed-off-by: YiPeng Chai Reviewed-by: Tao Zhou Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c | 67 +++++++++++++++++++++++++++++++++ 1 file changed, 67 insertions(+) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c index 7006a57277ef..0f2d765c4e2d 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c @@ -21,10 +21,13 @@ * */ +#include #include "amdgpu.h" #include "umc_v6_7.h" #define MAX_UMC_POISON_POLLING_TIME_SYNC 20 //ms +#define MAX_UMC_HASH_STRING_SIZE 256 + static int amdgpu_umc_convert_error_address(struct amdgpu_device *adev, struct ras_err_data *err_data, uint64_t err_addr, uint32_t ch_inst, uint32_t umc_inst) @@ -446,3 +449,67 @@ int amdgpu_umc_update_ecc_status(struct amdgpu_device *adev, status, ipid, addr); return 0; } + +static int amdgpu_umc_uint64_cmp(const void *a, const void *b) +{ + uint64_t *addr_a = (uint64_t *)a; + uint64_t *addr_b = (uint64_t *)b; + + if (*addr_a > *addr_b) + return 1; + else if (*addr_a < *addr_b) + return -1; + else + return 0; +} + +/* Use string hash to avoid logging the same bad pages repeatedly */ +int amdgpu_umc_build_pages_hash(struct amdgpu_device *adev, + uint64_t *pfns, int len, uint64_t *val) +{ + struct amdgpu_ras *con = amdgpu_ras_get_context(adev); + char buf[MAX_UMC_HASH_STRING_SIZE] = {0}; + int offset = 0, i = 0; + uint64_t hash_val; + + if (!pfns || !len) + return -EINVAL; + + sort(pfns, len, sizeof(uint64_t), amdgpu_umc_uint64_cmp, NULL); + + for (i = 0; i < len; i++) + offset += snprintf(&buf[offset], sizeof(buf) - offset, "%llx", pfns[i]); + + hash_val = siphash(buf, offset, &con->umc_ecc_log.ecc_key); + + *val = hash_val; + + return 0; +} + +int amdgpu_umc_logs_ecc_err(struct amdgpu_device *adev, + struct radix_tree_root *ecc_tree, struct ras_ecc_err *ecc_err) +{ + struct amdgpu_ras *con = amdgpu_ras_get_context(adev); + struct ras_ecc_log_info *ecc_log; + int ret; + + ecc_log = &con->umc_ecc_log; + + mutex_lock(&ecc_log->lock); + ret = radix_tree_insert(ecc_tree, ecc_err->hash_index, ecc_err); + if (!ret) { + struct ras_err_pages *err_pages = &ecc_err->err_pages; + int i; + + /* Reserve memory */ + for (i = 0; i < err_pages->count; i++) + amdgpu_ras_reserve_page(adev, err_pages->pfn[i]); + + radix_tree_tag_set(ecc_tree, + ecc_err->hash_index, UMC_ECC_NEW_DETECTED_TAG); + } + mutex_unlock(&ecc_log->lock); + + return ret; +} -- cgit From 2cf8e50ec381e6a6be3835a421f279d88fcb5ba4 Mon Sep 17 00:00:00 2001 From: YiPeng Chai Date: Mon, 22 Apr 2024 17:38:54 +0800 Subject: drm/amdgpu: Add delay work to retire bad pages Add delay work to retire bad pages. Signed-off-by: YiPeng Chai Reviewed-by: Tao Zhou Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c index 0f2d765c4e2d..2bd88218c20e 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c @@ -89,7 +89,7 @@ out_fini_err_data: return ret; } -static void amdgpu_umc_handle_bad_pages(struct amdgpu_device *adev, +void amdgpu_umc_handle_bad_pages(struct amdgpu_device *adev, void *ras_error_status) { struct ras_err_data *err_data = (struct ras_err_data *)ras_error_status; -- cgit From e74313be5a71df63e307ad98b6ab202b8a222817 Mon Sep 17 00:00:00 2001 From: YiPeng Chai Date: Fri, 22 Mar 2024 12:06:01 +0800 Subject: drm/amdgpu: add condition check for amdgpu_umc_fill_error_record Add condition check for amdgpu_umc_fill_error_record. Signed-off-by: YiPeng Chai Reviewed-by: Tao Zhou Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c | 20 +++++++++++++++++--- 1 file changed, 17 insertions(+), 3 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c index 2bd88218c20e..dcda3d24bee3 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c @@ -66,6 +66,8 @@ int amdgpu_umc_page_retirement_mca(struct amdgpu_device *adev, goto out_fini_err_data; } + err_data.err_addr_len = adev->umc.max_ras_err_cnt_per_query; + /* * Translate UMC channel address to Physical address */ @@ -121,6 +123,8 @@ void amdgpu_umc_handle_bad_pages(struct amdgpu_device *adev, if(!err_data->err_addr) dev_warn(adev->dev, "Failed to alloc memory for " "umc error address record!\n"); + else + err_data->err_addr_len = adev->umc.max_ras_err_cnt_per_query; /* umc query_ras_error_address is also responsible for clearing * error status @@ -146,6 +150,8 @@ void amdgpu_umc_handle_bad_pages(struct amdgpu_device *adev, if(!err_data->err_addr) dev_warn(adev->dev, "Failed to alloc memory for " "umc error address record!\n"); + else + err_data->err_addr_len = adev->umc.max_ras_err_cnt_per_query; /* umc query_ras_error_address is also responsible for clearing * error status @@ -389,14 +395,20 @@ int amdgpu_umc_process_ecc_irq(struct amdgpu_device *adev, return 0; } -void amdgpu_umc_fill_error_record(struct ras_err_data *err_data, +int amdgpu_umc_fill_error_record(struct ras_err_data *err_data, uint64_t err_addr, uint64_t retired_page, uint32_t channel_index, uint32_t umc_inst) { - struct eeprom_table_record *err_rec = - &err_data->err_addr[err_data->err_addr_cnt]; + struct eeprom_table_record *err_rec; + + if (!err_data || + !err_data->err_addr || + (err_data->err_addr_cnt >= err_data->err_addr_len)) + return -EINVAL; + + err_rec = &err_data->err_addr[err_data->err_addr_cnt]; err_rec->address = err_addr; /* page frame address is saved */ @@ -408,6 +420,8 @@ void amdgpu_umc_fill_error_record(struct ras_err_data *err_data, err_rec->mcumc_id = umc_inst; err_data->err_addr_cnt++; + + return 0; } int amdgpu_umc_loop_channels(struct amdgpu_device *adev, -- cgit From bfa579b38b865879223f61f6ae295c939d4f7d11 Mon Sep 17 00:00:00 2001 From: YiPeng Chai Date: Mon, 22 Apr 2024 17:40:03 +0800 Subject: drm/amdgpu: prepare to handle pasid poison consumption Prepare to handle pasid poison consumption. Signed-off-by: YiPeng Chai Reviewed-by: Tao Zhou Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c | 20 +++++++++++++------- 1 file changed, 13 insertions(+), 7 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c index dcda3d24bee3..8ebbca9e2e22 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c @@ -252,8 +252,9 @@ int amdgpu_umc_bad_page_polling_timeout(struct amdgpu_device *adev, return 0; } -int amdgpu_umc_poison_handler(struct amdgpu_device *adev, - enum amdgpu_ras_block block, uint32_t reset) +int amdgpu_umc_pasid_poison_handler(struct amdgpu_device *adev, + enum amdgpu_ras_block block, uint16_t pasid, + pasid_notify pasid_fn, void *data, uint32_t reset) { int ret = AMDGPU_RAS_SUCCESS; @@ -291,16 +292,14 @@ int amdgpu_umc_poison_handler(struct amdgpu_device *adev, amdgpu_ras_error_data_fini(&err_data); } else { - if (reset) { - amdgpu_umc_bad_page_polling_timeout(adev, - reset, MAX_UMC_POISON_POLLING_TIME_SYNC); - } else { struct amdgpu_ras *con = amdgpu_ras_get_context(adev); + amdgpu_ras_put_poison_req(adev, + block, pasid, pasid_fn, data, reset); + atomic_inc(&con->page_retirement_req_cnt); wake_up(&con->page_retirement_wq); - } } } else { if (adev->virt.ops && adev->virt.ops->ras_poison_handler) @@ -313,6 +312,13 @@ int amdgpu_umc_poison_handler(struct amdgpu_device *adev, return ret; } +int amdgpu_umc_poison_handler(struct amdgpu_device *adev, + enum amdgpu_ras_block block, uint32_t reset) +{ + return amdgpu_umc_pasid_poison_handler(adev, + block, 0, NULL, NULL, reset); +} + int amdgpu_umc_process_ras_data_cb(struct amdgpu_device *adev, void *ras_error_status, struct amdgpu_iv_entry *entry) -- cgit From 506c245f3f1cd989cb89811a7f06e04ff8813a0d Mon Sep 17 00:00:00 2001 From: Bob Zhou Date: Tue, 23 Apr 2024 13:32:39 +0800 Subject: drm/amdgpu: fix double free err_addr pointer warnings In amdgpu_umc_bad_page_polling_timeout, the amdgpu_umc_handle_bad_pages will be run many times so that double free err_addr in some special case. So set the err_addr to NULL to avoid the warnings. Signed-off-by: Bob Zhou Acked-by: Alex Deucher Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c | 1 + 1 file changed, 1 insertion(+) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c index 8ebbca9e2e22..540e0f066b26 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c @@ -179,6 +179,7 @@ void amdgpu_umc_handle_bad_pages(struct amdgpu_device *adev, } kfree(err_data->err_addr); + err_data->err_addr = NULL; mutex_unlock(&con->page_retirement_lock); } -- cgit